Loading summary
Transcript1 lines
- [00:00]
A
Welcome to the podcast. I lost my voice over the weekend so my voice is a little bit gone. We're, we're making a bit of a comeback. I'll try to talk quiet and boost the audio afterwards. I had a pretty busy weekend with my voice being lost. I ended up actually having to give a talk at church so that was a lot of fun. Whispering into microphone and also I went very viral on LinkedIn and Reddit for just a crap post that I was put out there was really funny but I the the numbers on it are hilarious and also the topic is AI related so I'll be covering that along with Meta and Andrell's AI Glasses collaboration, some bug bounty AI slop struggles that the industry is facing a huge energy merger with Next Era and Dominion for $66 billion. Amazon, Alexa plus doing AI podcasts, some companies raising money. There's a lot going on on the podcast today, so let's get into it. The first thing I want to talk about is kind of fun and funny. To kick the week off, I posted over on LinkedIn as a joke. I had my son just took a picture of me sitting on the back of my truck with a laptop on each of my knees. I recently got another laptop because the one I currently have wasn't powerful enough. So I let my wife have it. I actually stole it from her a long time ago. So I was just giving it back to her and got a new one. But anyways I had a laptop on each knee and I said just bought a second MacBook Air so I can now have two Claude Mac subscriptions running at once. Anyways, obviously this is a joke. I was like let me walk you through the math and then I go I proceed to say that a Claude Max plan is worth $11,000 a month in raw API compute and basically said that because I kept hitting my token limit. You know, it's so subsidized. You're just making so much money you have to have this and then this. The this is the part that I think triggered a lot of people. I said instant ROI math. The cost is $1,800 one time unlocked capacity is $11,000 a month. The payback period is under a week. Downtime eliminated 100% products in active development 13x this this then this is the line that I think triggered people. I said if you're still on one subscription in 2026, respectfully you're not serious. Anyways, this was a joke but it blew up and I ended up having to 250 people or actually 336 comments on this post. It had about a hundred thousand impressions on LinkedIn and then it got screenshotted and shared over to Reddit. On the LinkedIn lunatics Reddit where someone said I can't with these people. It had 1300 upvotes, 238 comments. Most of them are just people being like, hey, should we tell them about virtual machines? And then you know, mostly everyone just trying to scroll through my entire work history. Every company I've worked at saying everything I've ever done in my life is terrible. That was kind of the LinkedIn side of it over on, or so that was on the Reddit side of it over on LinkedIn it was pretty funny. There's probably a 50, 50 of people that were like haha, this is funny. And then probably 50% of the people that were trying to explain to me why running two laptops with two cloud max subscriptions was a bad idea and I could just run two clouds quad max subscriptions on one account, right? Like you get rate limited, just make a second account log into that. Anyways, it was pretty funny. Every single person that commented I replied to them mostly just doubling down and saying they obviously didn't understand the thermal envelope of my computer or I don't know, some other funny overly technical thing. Anyways, the results of it was actually pretty impressive to me on Google just from that I had 5,400 people search for the name of my company AI Box and go to my website on the first day. Google search console's a couple days behind so I only really know what happened on that first day. But it went pre viral the next day as well. So might be somewhere like 10,000 Google searches for my company. Because in all the screenshots specifically on Reddit it like it had my picture, my name on it. But actually I did have a really good day and we did hit our all time high in monthly recurring revenue and annual recurring revenue that day. So like we are uh, we are jumping up on things which is pretty, pretty funny all from that post. I did a follow up post today about it where I now have four laptops, but I think not as many people think it's real. So they're not quite as triggered by that one. Okay, let's talk about what's going on with Meta because I think this is a wild story happening right now over on Meta and Anduril. They both have just shared a bunch of new details about their augmented reality headset. So they're kind of Prototyping this right now with the U.S. army. But basically how this piece of hardware works is that a soldier can look at a target, they can tap their temple, and then they can talk in, you know, plain language. And the glasses are going to recommend a course of action. So it could include things like sending a nearby drone strike that still has to be proved approved up the normal chain of command. So it's not like AI is running the whole thing, but it is helping to, you know, be the interface. Right. It's kind of like having Siri or Alexa or something built into this. The contract is for 159 million doll for the prototyping phase of the Army's soldier born mission command program. And it's not just a one off. The army actually announced separately in March that it's spending $20 billion to integrate Anduril's Lattice software. That is the engine that pulls every army sensor and platform into one battlefield picture. And that's going to be happening across I think all of its infrastructure. So the glasses are kind of the user interface interface on the front end of this kind of 20 billion software backbone. And I think that changes basically the conversation. Right. This isn't just like a cool prototype. Um, Anduril is really becoming part of the platform layer for AI enabled warfare, which is very interesting. Um, something I think not a lot of people are talking about right now though is that these glasses are testing three different LLMs as the speech to text command layer. So they're using Gemini Llama and Claude Anthropic is not betting on any single AI company. Obviously Claude has so much beef with the US government. Just that's a new story in itself that this is included in there. Um, and anyways, Andrew is then going to pick whatever model handles military doctrine, classified content and edge device latency best. So they're testing them all out, not ruling Claude out, but they're testing out all the models. Whoever does it best is going to kind of get the contractor to get pulled into this tool. So for this whole project, components apparently started arriving in March and that's because of a bunch of federal contracting rules. All of those parts had to come from supply chains that didn't rely on Chinese companies. Meta's, you know, commercial ray Bans that they have, those can come from China, that doesn't really matter. But when you make these, you know, something for the U.S. army, they don't want to be, to be reliant on China. So I think this is also a bit of a domestic manufacturing build that's happening right now and Meta is building and kind of the, the displays and the wave guides. Andrel does a lot of the system integrations. Of course. The. Another funny part of this is that Meta and Andrell working together is interesting considering Palmer Lucky started Oculus, got acquired by Meta and then got kicked out of Meta due to a lot of activists like Jason Calacanis. But now he's back working with them, so that's kind of funny. He posted on X something that's like people praying for my downfall need to pray harder. Anyways, so he obviously has a chip on his shoulder, but that's good because he's pushing and trying to make some, you know, useful software. So there's also the self funded Andrew side of this whole thing project called Eagle Eye. It's going to be built into the helmets and headsets. The army hasn't asked for this, but Andrew says that they're gonna get. They're. They're adding this. Anyways, the next story I want to cover is actually something I've been hearing about for a while, but apparently the problem has been getting worse. So I feel like it's worth bringing up again. But apparently COR corporate bug bounty programs are basically just getting completely flooded by AI generated submissions. And so some of them are just completely shutting down. Some of these big companies that had bug bounty programs, they're just shutting them down, which is kind of a bummer. Bug crowd whose customers include OpenAI T Mobile, Motorola, they said that report reports more than quadrupled over three weeks stretch in March and most of them were fake. So that's kind of the hard thing. People will go in and tell Claude to go scrape all the code, try to find bugs, um, and then submit kind of fake reports. And it's really hard to go weed through all of them. HackerOne, which runs bounties for Goldman Sachs, Google and the Department of Defense, they logged a 76% jump in submissions in a year. And what's interesting is out of that, right, they saw like a 76% jump. But only I think, you know, 1, 1 out of 4 was an actual vulnerability, was a real one. So vulnerabilities are still flat at around 25% and most of the leads that they're getting go nowhere. So this is a really big problem that's happening right now. We're liter having to use and rely on AI to filter these AI bug reports. Okay, let's talk about the huge merger happening between Next Era and Dominion. $66.8 billion Next Era just agreed to acquire Dominion. This is one of the biggest US utility transactions ever announced. And the deal is basically being framed openly as a bet on AI driven power. Demand Routers was reporting on this and apparently the combined company would be one of the largest investor owned utilities in the entire country. The why this isn't so much about utilities as it is about AI is because Dominion's service territory, it covers Virginia, which hosts the densest concentration of hyperscaler data centers in the world. If you've ever been on like AWS or I don't know, like many like VPNs or, or there's just like a lot of, if you've done anything with where you're picking like you, where your server location is, it's always like do you want it in Virginia or do you want it in like, you know, on the east coast, in like California? Like you usually have like a couple of choices, but there's always the Virginia option because there's just so many data centers there right now. Ludon county alone has a huge chunk of global Internet traffic going through it in Virginia. The reason I personally think this is a really big deal is because the bottleneck for these AI build outs is not just GPUs anymore because that was kind of the problem before it was how you could get your hands on enough GPUs from Nvidia Etc. Now I think it's not even just the capital, it's basically the megawatts and it's a lot of the interconnection queues. So Dominion's Virginia interconnect queue has stretched into multi year weights. There is a, you know, putting together a $66.8 billion utility tie up. It only makes sense if you believe that AI electricity demand is structural, it's not cyclical. Right. This isn't just a big bubble. So this would make a lot of sense. And then if that's the case, then owning a lot of this regulated load serving capacity is basically the most defensible position in the AI value chain. So Nextera is making that bet. They're putting, they're pretty clear about this. You know, they're betting like hey, it's going to be hard for people to build more of these data centers in Virginia. We're basically going to be powering all of them. There's also another interesting part of this which is that the state commissions in Virginia, Florida and FERC at the federal level, all of them have to sign off. So approvals on deals this size historically take 12 to 24 months, sometimes longer. Definitely. There's a lot of interventions and a lot of, you know, regulation that goes through. A lot of people are going to push hard to for rate payer protections. But if all of this clears, if this merger acquisition, I guess you could say is actually going to happen, I think the most valuable real estate in AI isn't just going to be the model weights, right? It's, it's basically the interconnection rates behind them. Next up, Amazon's Alexa plus can now generate AI podcasts on demand. My job is officially at risk. You basically can tell it a topic and it's going to show you the outline and then they'll have two AI hosts that are kind of talking about it. You can adjust the length and if you minutes later doing all of this, your Echo show pings and you're ready to listen to the full episode. Routers talked about this and a bunch of other publishers as well, but I think the format's not new. Go Google's Notebook lm. We've been talking about it a lot on the show, but you can do this. You can put a PDF or you can put your, you know, your schoolwork or a paper or some sort of work report into Notebook LM and it'll have like a podcast where it's two people talking about it. A lot of people find this more digestible than just trying to get through the content and read it yourself. But of course, Notebook lm, you got to go to the we website and upload it, whereas the Alexa, you can just talk to it out loud, tell it the topic and it can put something together for you. I could be completely wrong on this bet, but when I first started this podcast, basically my assumption was because there was a lot of AI tools out there, I wasn't going to clone my voice and just dump in AI news to be read because people probably wanted to hear it from a real person. Now you can see the downsides of this theory because right now my voice is lost. And so maybe people think I sound terrible, I don't know. Or maybe it's just harder for me to talk when my voice is lost. I mean, luckily it happened on like Friday, so I had Saturday and suddenly to slightly recover. But you can see there are some pros to having an AI voice host, but you kind of miss the, the anecdotes from the person talking, which a lot of people like. Okay, the last story I want to talk about is Lenin ar. This is a Korean optics startup. They just closed an $18.5 million round of funding from the Korean Development bank and from Lottie Ventures. So they've raised total $41.7 million. They actually have a IPO planned for next year in Seoul. So what the heck does this company do? Why is show? This is what was really shocking to me. Global AI glasses shipments hit 8.7 million units last year. That is a 300% jump year over year and Omdia is projecting 15 million units this year. So Lennon, AR, this company I've been talking about, they don't just make the glasses, they actually make the optical module that is the tiny lens component that projects images onto your field of view. Their CEO is Jaehyuk Kim and he said that the optics module is the hardest part to get right because it has to be thin, light and power efficient at the same time. And every glasses maker has to source one. So they have something called pen tilt design which is going to direct light precisely towards the eye instead of scattering it across the lens like a dominant waveguide approach. This is going to throw away most of the photons and it's going to drain the battery a lot faster. So they're already shipping inside a Swiss AR motorcycle helmet from iegas Rider. It's a spin out of ETH Zurich's Computer Vision Lab and that anchors navigation arrows to the actual road at about 160 km an hour, which is a super cool use case. They have a bunch of other big customers. One of them is called Dynabook in Japan. LG was an earlier backer and is reportedly building their own AI glasses now, so they're kind of working with them. Personally, I love to talk about the companies behind a lot of the trends like AR glasses and the companies that are winning because of that. So this is a really fascinating story to me guys. Thank you so much for tuning into the podcast. If you enjoyed the episode today, make sure to leave us a rating and review wherever you get your podcasts. Hopefully my voice gets a little bit better. Maybe I'll splice in some podcasts from my AI chat or AI Applied podcast if if I'm not recovering too quickly. But yeah, thanks so much guys for tuning in. Make sure to check out AI box AI it is 8.99amonth to get access to over 80 AI models all in one place. And also if you want to, by the way, for the LinkedIn post that I did, I'll leave a link in the show notes to that. It's funny, but if you want to check out all of these different podcasts, I talk about and get a more in depth view as either a newsletter or as or as an article that you can read. Go check out aichatdaily. Com. I'll also leave a link in the description to that. All right, catch you guys all in the next episode.